Skip to content

[Offload][AMDGPU][NFC] Improve message for warning non-XNACK runs#2345

Merged
ronlieb merged 2 commits into
amd-stagingfrom
improve-apu-error-msg
Apr 30, 2026
Merged

[Offload][AMDGPU][NFC] Improve message for warning non-XNACK runs#2345
ronlieb merged 2 commits into
amd-stagingfrom
improve-apu-error-msg

Conversation

@Kewen12
Copy link
Copy Markdown

@Kewen12 Kewen12 commented Apr 28, 2026

This PR is to provide detailed warning message for guiding users when running USM kernels on the non-XNACK APUs.

@Kewen12 Kewen12 requested review from jplehr and ronlieb April 28, 2026 22:23
@z1-cciauto
Copy link
Copy Markdown
Collaborator

Copy link
Copy Markdown

@jplehr jplehr left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think that change makes sense to improve communication to users.

"re-run with HSA_XNACK=1. If your device does not support XNACK, "
"remove USM pragma and use map clauses instead. "
"Set OMPX_EAGER_ZERO_COPY_MAPS=1 for optimal zero-copy "
"performance on non-XNACK APUs.");
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
"performance on non-XNACK APUs.");
"performance on non-XNACK shared-memory devices.");

I don't know whether we should use the term APU here. I don't have a strong opinion either.

@z1-cciauto
Copy link
Copy Markdown
Collaborator

@Kewen12
Copy link
Copy Markdown
Author

Kewen12 commented Apr 29, 2026

Updated the USM tests to address the CI failure: ROCm/aomp#2206

@ronlieb
Copy link
Copy Markdown
Collaborator

ronlieb commented Apr 30, 2026

!PSDB

@z1-cciauto
Copy link
Copy Markdown
Collaborator

@ronlieb ronlieb merged commit 1300277 into amd-staging Apr 30, 2026
49 of 51 checks passed
@ronlieb ronlieb deleted the improve-apu-error-msg branch April 30, 2026 20:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants